evaluation of urban mobility using surveillance cameras · 2017. 1. 18. · evaluation of urban...

doi: 10.1016/j.procs.2015.11.042

Evaluation of urban mobility using surveillance cameras

Alexey V. Kurilkin, Oksana O. Vyatkina, Sergey A. Mityagin, Sergey V. Ivanov

ITMO University Saint-Petersburg, Russia.

[email protected]

Abstract Urban mobility is an important part of many studies related to the planning of large urban areas. This paper describes an approach for the evaluation of urban mobility as geosocial dynamics on the territory with high population density on the basis of surveillance cameras. We examine various methods of image and video processing for the evaluation of people flow. As a result, we propose a fast processing and low-cost method for general purpose cameras for the estimation of the number of moving people at a particular point in the city. A case study for some locations in St.Petersburg is considered. Keywords: social dynamics, surveillance cameras, image processing, urban mobility, mobile phones

1 Introduction In a broad sense, urban mobility and geo-social dynamics includes processes of interaction of

social and geographical space (Holderness,2014; Chumakova,2011). Features of the geographical space of the specific urban territory, combined with high population density determines the need for new approaches to the management of such areas to achieve the operational and strategic objectives.

In the present urban mobility and work geo-social dynamics is considered as the movement of the masses of people in the city. Research on geo-social dynamics in large urban areas refers to the area in which carrying out effective field experiment is difficult or impossible. For that reason, the use of video surveillance cameras in the fixed points is a reasonable solution for the evaluation of the overall dynamics of the population in the city. Such assessments are needed for the operational management of the city and security, transportation access, spatial planning and other tasks (Steenbruggen,2013; Berlingerio,2013; Min,2010).

Simulation of urban mobility in large areas in this study can be reduced to the evaluation of the evolution of the spatial distribution of the number of people on different time scales.

Procedia Computer Science

Volume 66, 2015, Pages 364–371

YSC 2015. 4th International Young Scientists Conference onComputational Science

364 Selection and peer-review under responsibility of the Scientific Programme Committee of YSC 2015c© The Authors. Published by Elsevier B.V.

http://crossmark.crossref.org/dialog/?doi=10.1016/j.procs.2015.11.042&domain=pdf

2.1 Formal description and related works In this study we consider the spatial distribution of the number of people as an estimation the

number of people in a geographical area ijC bounded by a square with a side a at a specific time .

The dynamics of the spatial distribution may be considered in terms of changes in the number of

people in the area )(ijC , either in terms of movement of individuals between geographic areas. In a study (Song,2010), the authors consider the example of the analysis and prediction of the

movement of people according to the movement of mobile subscribers in Côte d'Ivoire. In this work it was shown that the theoretical maximum predictability of trajectories was higher than 88% with Markov chain models. The accuracy of the prediction was 97% for fixed trajectories and 95% for non-stationary trajectories. It was also noted that the degree of predictability of territorial dynamics strongly depends on the historical behavior, and the maximum predictability may vary with different goals.

At the same time, after the revision of the results a factor of the impact of spatial partitioning of territories was added (Smith,2014). Thus, changes in behavior happen fairly slowly, and movement have a high degree of spatial and temporal regularity and a likelihood to return in frequently visited places is great. People follow simple reproducible patterns of movement. The similarity in movement patterns will also affect all the phenomena that depend on mobility (González,2009). While most researchers analyze the mobility in long-term range, mobility in the scale of the one day remains poorly understood. Obviously, daily mobility patterns are characterized by the fact that the starting point is normally coincides with the end point with a fixed set of intermediate points, and a physical location in this context does not play a significant role.

Using the concept of network motifs authors in (Schneider,2013) identified 17 unique networks with simple rules that are present in everyday activity. According to the study, these 17 motifs was sufficient to cover 90% of the population in different countries. Usually, each person has a motif that persists for several months.

The social environment also contributes to the mobility of the population. Deviations from standard patterns of behavior are often caused by the influence of the social relations.

While short trips are periodic in the spatial and temporal dimension, long trip greatly influenced by social connections.

Social interactions can explain 10 to 30 percent of the movements while the share of periodic movements takes from 50 to 70 percent (Cho,2011). Mobility is strongly dependent on social contacts. Users with a large number of different contacts eager to explore new places, and there is a strong correlation between the strength of the social relations and the sustainability of commuting (Toole,2015).

By analyzing the mobility of the population in a particular area, we identify places of interest and a sequence of movements that are typical for tourists. If we consider how individual visit a particular location as a link with a certain weight, it is possible to use models based on graphs.

An important element for the understanding of the territorial mobility is the correspondence matrix in different scales (Çolak,2015; Simini,2013). The obtained matrices are easy to compare demographic and transportation data. Mobility patterns derived from correspondence matrices, as a rule, are universal and can be applied to any territory, if we have the respective data.

2.2 Source data for urban mobility As the initial data on the territorial mobility we consider data from a variety of sources, the most

common of which are information about official registrations, data from mobile operators, and geosocial data from online media. However, these sources do not provide operational information on the mobility of the population and are not well suited for the simulation of daily activities.

Evaluation of urban mobility using surveillance cameras Kurilkin et al.

365

The data on the official registration allows to estimate the overall population density in the area and possibly classify areas for the different intended purpose. However, this information is not sufficient for the intraday analysis of population mobility. Current sources of information are characterized by some restrictions, such as low level of accuracy of official statistics, the lack of data on the people commuting inside the day, inability to obtain online data for operational analysis.

Data from mobile operators on the other hand, do not have these disadvantages, and thanks to the widespread use of mobile phones can deliver detailed information about the population activity without special efforts. A new source of information makes it possible to supplement and clarify the existing data on the territorial mobility (Isaacman,2012).

The use of mobile operators and mobile positioning technology allows subscribers to discover new possibilities for the study of urban space. A new source of information can be applied in various fields, such as monitoring the population, traffic management, etc. Many European countries are interested in the data from mobile operators, in particular for an urgent task like migration and tourism in the European Union (Paraskevopoulos,2013; Dong,2015; Wesolowski,2009).

Figure 1 shows an example of comparing the distribution of the density of people from official statistics and data of one of the mobile operators.

a) b) c) d)

Figure 1. The density distribution of people in the Central District of St. Petersburg. a) information of official registration; b) data from mobile operator in 6.00; c) data on the mobile operator 14.00; d)

data mobile operator in 21.00 Data from mobile operators can be used to understand the behavior of individual and overall

mobility, as for ordinary urban situations and for different events. Cellular operators store huge amounts of information about subscribers. These data includes

personal information as passport data, tariff plan, mobile phone model, graph calls. The special kind of information is a location at a given moment, identified by the cell stations.

Each mobile phone searches for the nearest base station, listening to the broadcast for available cells. The mobile device defines several most satisfying channels with minimizing of energy costs and signal quality. All base stations are grouped. Belonging to a group is determined by the location. For the identification of the base station each group has a unique Location area code (LAC). Each of the cells has a unique number (CellID), which, mutually with LAC uniquely identifies the cell station, on which mobile device is working. Furthermore, a system determines the cell station sector (CellSector) and fixes the time during which the signal goes from the mobile device to the cell station (Timing Advance). Thanks to this feature it is possible to determine the distance to the base station. To improve the accuracy of the location of the mobile user a method that takes into account the Received Signal Strength (RSS) of the nearest cell station is used. With a triangulation from several cell station, it allows to calculate the position of the mobile user (Promnoi,2008). Thus, positioning accuracy depends on the density of the cell stations. Error in determining the coordinates is less in the city. In urban areas location accuracy is generally close to square with 250 meters sides. To further increase the accuracy and tracking of mobile users one can apply recovery trajectories algorithms based on cell stations data (Leontiadis,2014).


366

Thus, data from mobile operators cannot be fully utilized to obtain operational information on people mobility in the urban area because of some technical difficulties.

For these reasons, we consider surveillance cameras as the main source of operational data on the people mobility. This data source is more accessible and relatively accurate in comparison to other methods. We use only general purpose camera, and this allows extensive use of this method.

2.3 Evaluation of social dynamics on the basis of surveillance cameras at micro-level

Data from surveillance cameras allow to obtain a quick assessment of the number of people going under particular sector of the camera in short period. Thus, each camera plays a role of the sensor, counting the number of people walking on a street per unit time. Thus, our task is reduced to an estimation of number of people going through the sector of surveillance camera.

When people are passing under the camera, they overlap each other, and it confuses the identification of the particular person in the image or video. Therefore, in this study we used the method based on the regression model to estimate the number of people going into camera view. The method also allows to get a good performance on video processing and can be applied to both a video stream and a video files. It is significant that the method can be used for low-resolution video images because it is not associated with the identification of specific objects.

Figure 3. The main view form the camera with region of interest

For each camera, we perform pre-setting of the region of interest in accordance with the visible

perspective (Figure 2). In the foreground and background we set the maximum size of a visible person in pixels. This operation is necessary for the correct calculation taking into account the perspective distortion. To calculate the number of people input video stream from each camera is divided into a sequence of frames. These frames we convert to black and white view and produce pre-processing with filters to improve image characteristics. By subtracting the previous frame from the current one we have the foreground image, which is passed through a threshold filter, highlighting thereby moving objects. For the pixels of the resulting image we make a perspective correction. The number of pixels per line is multiplied by the weight that decrease to the upper limit.


367

a) b)

Figure 3. Counting pixels in the image: a) the original image, b) after the merging of the pixels (morphology operation)

To correct perspective, we use the approach described in (Ma,2004). The size of the object is

changed by a linear function of y coordinates of the image. To make objects of the same size, we used the following equation

, (1) where is the size of the object, is size of the object at the bottom of the image, and is the ratio of the size of the object. Aspect ratio is calculated for each image line with the formula

, (2)

where determines the line, which is the point of convergence (determined by a visible image of one object in different parts of the image), is a distance from to current line of the image, is the distance from to low bottom of the image. After perspective correction, the resulting number of pixels is calculated by the equation

, (3) where N is the total number of pixels, Y – number of lines in the image, is the number of pixels per line y, is aspect ratio for a given line.

Figure 4. The dependence of the number of people on the number of pixels in the image, calculated in different algorithms of the perspectives correction with regression model


368

Calculation of the moving pixels in the image with the perspective is accomplished twice: the first in the original image (Figure 3a), and the second time in the image after application of morphology (Figure 3b). The use the morphology operation is necessary for the identification of people standing in the frame or moving at low speed. These numbers are input to the regression model, which is used for estimating the number of people in the frame.

Each camera requires a preliminary calibration. It lies in manual counting the number of people in the field of view of the camera and comparison with the number of pixels obtained after perspective correction and after the application of the morphology (pixels merging). Using these data we build a regression model that allows us to estimate the number of people in the frame. Thus, in Figure 4 and Figure 5, we show the results of the calibration the camera, located in the Central district of St. Petersburg. It is easy to observe that in the case without perspective correction the number of pixels leads to a greater spread of people. The similarity of the graphs with different methods of calculation with perspective correction can be explained by a small number of people who were in the field of the camera.

Figure 5. The dependence of the number of people on the number of pixels in the image

(calculation without perspective correction) with regression model

3 Case studies Using data from mobile operators and surveillance cameras we can automatically identify and

predict city events based on behavioral patterns. It allows to determine a social response to public events. It helps to allocate crowds and simulate its behavior. Many cities have already introduced systems that allow to predict the occurrence of zones with a high number of crimes (Bogomolov,2014). Information on the distribution of the people in the urban area, as well as on the mobility can improve building of strategies in the case of serious epidemics or emergency. During the clustering of data, we can identify various urban areas with their characteristic and dynamics of the


369

population, such as residential areas, commercial, industrial, area parks. Another type of research is related to the detection of city events, which is possible with sufficiently high spatial and temporal accuracy in case of the presence of a sufficient quantity of historical data. It is also possible to do clustering and search for similar events in the historical data. Clustering is also used to identify different social groups and significant events. Important to notice that data can be combined with a variety of sources, for example, we can try to identify a tourist by his messages and online photos with geolocation (Girardin,2009). There is also the opportunity to assess the attractiveness of the territory on the basis of social networks. Combining this information with the behavior patterns obtained by analyzing the data, we can more deeply estimate the attractiveness of urban space for various population groups (Kling,2012). The combination of these social relations with the individual pathways provides a new look at the tourist routes or routes of individual groups. For systems of this class, we had a number of requirements, such as visualization of large amounts of data, the ability to perform time series analysis in the spatial dimension, and taking into account seasonal factor and changes in the population dynamics.

4 Conclusions In this paper we considered the issues of urban mobility and geosocial dynamics in areas with high

population density and using different input data. Due the existing restrictions almost none of single data source can not fully be used as the main information resource to solve this problem. The most significant limitation is the ability to estimate the current distribution of people on the territory. In this paper we offer an approach to solving this problem using general purpose surveillance cameras with image processing and regression model. This method was used to assess the number of people in the Central district of St. Petersburg. Preliminary results hint that the most significant problems for the full-fledged implementation of the proposed method is the accuracy of estimating the number of people going in the given sector of the camera. The second important thing is how to provide the comparability of results from multiple surveillance cameras with different characteristics. These challenges are the main directions for future works.

Acknowledgements. This work was financially supported by the Russian Foundation for Basic

Research, Grant 15-29-07034 "Technology of distributed cloud computing for the simulation of a large city".

References 1. Holderness, T. (2014). Geosocial intelligence. IEEE Technology and Society Magazine, 33 (1),

17-18. 2. Chumakova S.V. (2011) Model of interaction between of geographical and social space. / //

European Social Science Journal – 2011, 271-273 3. Steenbruggen J. et al. (2013) Mobile phone data from GSM networks for traffic parameter and

urban spatial pattern assessment: a review of applications and opportunities //GeoJournal. – 2013. – Т. 78. – №. 2. – С. 223-243

4. Berlingerio M. et al. (2013) AllAboard: a system for exploring urban mobility and optimizing public transport using cellphone data //Machine learning and knowledge discovery in databases. – Springer Berlin Heidelberg, 2013. – С. 663-666.


370

5. Min W., Wynter L. (2010) Real-time road traffic prediction with spatio-temporal correlations //Transportation Research Part C: Emerging Technologies. – 2011. – Т. 19. – №. 4. – С. 606-616.

6. Song C. et al. Limits of predictability in human mobility //Science. – 2010. – Т. 327. – №. 5968. – С. 1018-1021

7. Smith G. et al. (2014) A refined limit on the predictability of human mobility //Pervasive Computing and Communications (PerCom), 2014 IEEE International Conference on. – IEEE, 2014. – С. 88-94.

8. González M. C., Hidalgo C. A., Barabási A. L. (2009) Understanding individual human mobility patterns //Nature. – 2009. – Т. 458. – №. 7235. – С. 238-238.

9. Schneider C. M. et al. (2013) Unravelling daily human mobility motifs //Journal of The Royal Society Interface. – 2013. – Т. 10. – №. 84. – С. 20130246.

10. Cho E., Myers S. A., Leskovec J. (2011) Friendship and mobility: user movement in location-based social networks //Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining. – ACM, 2011. – С. 1082-1090.

11. Toole J. L. et al. (2015) Coupling human mobility and social ties //Journal of The Royal Society Interface. – 2015. – Т. 12. – №. 105. – С. 20141128.

12. Çolak S. et al. (2015) Analyzing cell phone location data for urban travel: current methods, limitations, and opportunities //Transportation Research Board 94th Annual Meeting. – 2015. – №. 15-5279.

13. Simini F. et al. (2012) A universal model for mobility and migration patterns //Nature. – 2012. – Т. 484. – №. 7392. – С. 96-100.

14. Isaacman S. et al. (2012) Human mobility modeling at metropolitan scales //Proceedings of the 10th international conference on Mobile systems, applications, and services. – ACM, 2012. – С. 239-252.

15. Paraskevopoulos P. et al. (2013) Identification and characterization of human behavior patterns from mobile phone data //Proc. of NetMob. – 2013.

16. Dong Y. et al. (2015) Inferring Unusual Crowd Events From Mobile Phone Call Detail Records //arXiv preprint arXiv:1504.03643. – 2015.

17. Wesolowski A. P., Eagle N. (2009) Inferring Human Dynamics in Slums Using Mobile Phone Data. – Technical report, Santa Fe Institute, 2009.

18. Promnoi S., Tangamchit P., Pattara-Atikom W. (2008) Road traffic estimation based on position and velocity of a cellular phone //ITS Telecommunications, 2008. ITST 2008. 8th International Conference on. – IEEE, 2008. – С. 108-111.

19. Leontiadis I. et al. (2014) From Cells to Streets: Estimating Mobile Paths with Cellular-Side Data //Proceedings of the 10th ACM International on Conference on emerging Networking Experiments and Technologies. – ACM, 2014. – С. 121-132.

20. Bogomolov A. (2014) et al. Once upon a crime: Towards crime prediction from demographics and mobile data //Proceedings of the 16th International Conference on Multimodal Interaction. – ACM, 2014. – С. 427-434.

21. Girardin F. et al. (2009) Quantifying urban attractiveness from the distribution and density of digital footprints. – 2009.

22. Kling F., Pozdnoukhov A. (2012) When a city tells a story: urban topic analysis //Proceedings of the 20th International Conference on Advances in Geographic Information Systems. – ACM, 2012. – С. 482-485.

23. R. Ma, L. Li,W. Huang, and Q. Tian (2004) On pixel count based crowd density estimation for visual surveillance // in Proc. IEEE Conf. Cybern. Intell. Syst., 2004, pp. 170–173.


371

evaluation of urban mobility using surveillance cameras · 2017. 1. 18. · evaluation of urban...

Documents